Synthetic Data Generation in Healthcare Using GANs
Author Information
Author(s): Muhammed Halil Akpinar, Abdulkadir Sengur, Massimo Salvi, Silvia Seoni, Oliver Faust, Hasan Mir, Filippo Molinari, U. Rajendra Acharya
Hypothesis
This systematic review aims to provide a comprehensive overview of GAN-based synthetic data generation methods in healthcare, focusing on image and signal-based studies across various clinical domains.
Conclusion
The review highlights the potential of GANs in generating synthetic medical data, improving diagnostic accuracy, and addressing data scarcity in healthcare.
Supporting Evidence
- GANs can improve AI model performance by augmenting existing datasets with synthetic data.
- The review identifies promising data augmentation, anonymization, and multi-task learning results.
- Brain studies constitute the most researched clinical domain, followed by biomedical signals and cancer.
- Current limitations include the lack of standardized metrics and direct comparisons.
- Future directions include the development of no-reference metrics and enhanced interpretability.
Takeaway
This study looks at how computers can create fake medical data to help doctors and researchers when they don't have enough real data to work with.
Methodology
The review followed PRISMA guidelines and analyzed 72 relevant journal articles on GAN applications in healthcare.
Potential Biases
Potential bias due to the limited diversity of datasets and the reliance on specific GAN architectures.
Limitations
The lack of standardized metrics and benchmarks complicates the comparison of different GAN models and the assessment of their real-world impact.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website